diff --git a/benchmark/baseline/ci.json b/benchmark/baseline/ci.json
index 331d481..72b10ef 100644
--- a/benchmark/baseline/ci.json
+++ b/benchmark/baseline/ci.json
@@ -1,19 +1,19 @@
 {
-  "_comment": "Seeded from scaling run 20260618-181019 + footprint run 20260618-185748 on Intel i7-10700K 16-core. gateway_sha for footprint was 'unknown' (demo image built before SHA capture; will be real on next footprint build). Scaling gateway_sha: 8569f213e7a1e4b4fd2f699546eda9a11e78dcf1.",
-  "gateway_sha": "8569f213e7a1e4b4fd2f699546eda9a11e78dcf1",
+  "_comment": "Baseline from run 20260622-220119. gateway_sha bdcc4ed08008fcb7c37e6af9b2074831f4f5abd8 is the measured gateway commit for every lane in this run.",
+  "gateway_sha": "bdcc4ed08008fcb7c37e6af9b2074831f4f5abd8",
   "host": {
     "cpu_model": "Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz",
     "nproc": 16,
-    "mem_total_kb": 32824960
+    "mem_total_kb": 32824968
   },
   "footprint": {
-    "footprint.uss_kib_median": 92960.0,
-    "footprint.cpu_cores_median": 0.19535913385165726,
-    "footprint.num_threads_median": 53.0
+    "footprint.uss_kib_median": 54248.0,
+    "footprint.cpu_cores_median": 0.24552932830433832,
+    "footprint.num_threads_median": 31.0
   },
   "scaling": {
-    "scaling.exponent": 0.4556986360933056,
-    "scaling.ci_lo": 0.26119683252603965,
-    "scaling.ci_hi": 0.6502004396605715
+    "scaling.exponent": 0.2930011978650536,
+    "scaling.ci_lo": 0.03479966730245926,
+    "scaling.ci_hi": 0.551202728427648
   }
 }
diff --git a/benchmark/benchmark.py b/benchmark/benchmark.py
index 8c541f7..5ee139e 100644
--- a/benchmark/benchmark.py
+++ b/benchmark/benchmark.py
@@ -882,13 +882,37 @@ def cmd_update_baseline(args):
             run_meta = json.loads(meta_path.read_text())
             break
 
+    # gateway_sha can differ per lane (synthetic image vs demo image), so collect
+    # every lane's value and describe the real source instead of a generic guess.
+    lane_shas = {}
+    for lane_dir in sorted(run_dir.iterdir()):
+        meta_path = lane_dir / "run_metadata.json"
+        if lane_dir.is_dir() and meta_path.exists():
+            lane_shas[lane_dir.name] = json.loads(
+                meta_path.read_text()).get("gateway_sha", "unknown")
+    real_shas = sorted({s for s in lane_shas.values() if s != "unknown"})
+    gateway_sha = real_shas[0] if real_shas else "unknown"
+    if lane_shas and all(s != "unknown" for s in lane_shas.values()) and len(real_shas) == 1:
+        comment = (
+            f"Baseline from run {run_dir.name}. gateway_sha {gateway_sha} is the "
+            "measured gateway commit for every lane in this run."
+        )
+    elif lane_shas:
+        per_lane = ", ".join(f"{k}={v}" for k, v in sorted(lane_shas.items()))
+        comment = (
+            f"Baseline from run {run_dir.name}. gateway_sha per lane: {per_lane}. "
+            "Top-level gateway_sha is the first measured commit; 'unknown' means "
+            "that lane's image predated SHA capture."
+        )
+    else:
+        comment = (
+            f"Baseline from run {run_dir.name}. gateway_sha unknown "
+            "(no lane run_metadata.json found)."
+        )
+
     baseline: dict = {
-        "_comment": (
-            "Baseline seeded from run: " + run_dir.name +
-            ". gateway_sha may be 'unknown' for footprint lane if "
-            "the turtlebot3 demo image was built before SHA capture was added."
-        ),
-        "gateway_sha": run_meta.get("gateway_sha", "unknown"),
+        "_comment": comment,
+        "gateway_sha": gateway_sha,
         "host": {
             "cpu_model": run_meta.get("cpu_model", "unknown"),
             "nproc": run_meta.get("nproc", 0),